Well, it turns out I was wrong about the 2s banding. This _was_ garbage collection, and was cured by using -XX:+UseConcMarkSweepGC -XX:+UseParNewGC.
I still get the 5s banding, though, which is a puzzler. Edward. -----Original Message----- From: Kurt Overberg [mailto:[EMAIL PROTECTED] Sent: 31 July 2006 12:08 To: Tomcat Users List Subject: Re: Response time banding Do your 2 second requests happen at a particular time of day? In the early morning, when my database vacuum is running, my response times shoot way up (to around 5 seconds). Check to see what else is going on with your system during the times the page response times increase. /kurt On Jul 26, 2006, at 4:47 PM, Robert Harper wrote: > One thing to consider is that the times may not be totally in > synch. Some > threads pending a write to the log might get interrupted while another > thread runs. You may see log entries out of order. I have seen this > myself > and have added System.out.flush() to my logger. It helps but still > does not > ensure exact order. Make sure you go over the time stamps carefully > before > you pass too much judgment. > > Robert S. Harper > Information Access Technology, Inc. > > -----Original Message----- > From: Leon Rosenberg [mailto:[EMAIL PROTECTED] > Sent: Wednesday, July 26, 2006 2:35 PM > To: Tomcat Users List > Subject: Re: Response time banding > > I would measure at all layers during the request processing to > determine whether irregularities occur somewhere. > > Leon > > P.S. "the response times as measured by the access log" - don't you > have the proper tools for that? JMeter? > > On 7/26/06, Edward Hibbert <[EMAIL PROTECTED]> wrote: >> Odd one, this. I'm posting looking for off the wall suggestions. >> Though not too off the wall, please. >> >> We have an application running under Tomcat. If we analyse the >> response >> times as measured by the access log, then we see that for the same >> GET >> operation there is a distribution of response times. Well, you'd >> expect >> that. >> >> But while the bulk of response times are below about 0.25 seconds (in >> most cases considerably below), there is a banding effect where we >> get a >> cluster of response times around 2 seconds, and another cluster >> around 5 >> seconds, with little or no values in between. >> >> We've looked at garbage collection as the most likely source of this, >> and ruled it out. Has anyone seen anything similar, or got any >> bright >> ideas? >> >> Regards, >> >> Edward. >> >> > > --------------------------------------------------------------------- > To start a new topic, e-mail: users@tomcat.apache.org > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > > > > > > --------------------------------------------------------------------- > To start a new topic, e-mail: users@tomcat.apache.org > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > --------------------------------------------------------------------- To start a new topic, e-mail: users@tomcat.apache.org To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --------------------------------------------------------------------- To start a new topic, e-mail: users@tomcat.apache.org To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]